Cobot buyers should lock seven evaluation axes before opening a vendor catalog: payload margin (≥50% over the heaviest part), reach envelope, repeatability, ISO/TS 15066 safety functions, force-limiting thresholds, communication protocol, and cell-calibration tolerance [S2]. The Mecheri and Greene analytical hierarchy process (AHP) framework, published in <em>International Journal of Rapid Manufacturing</em> Vol.8 No.4 (2019), remains the only peer-reviewed multi-criteria decision methodology built specifically for collaborative robot selection [S2].
Specifications that look identical on a 6-axis cobot and a collaborative robot datasheet often diverge by 2-3× once the workcell model is calibrated against measured data, per Horváth, Erdős and Kemény (2021) [S3]. Treat the catalog repeatability number as a nominal value and budget for 0.05-0.1 mm of additional deviation once the arm is integrated with grippers, fixtures, and vision.
Payload, Reach, and Repeatability Triplet
The three mechanical axes that disqualify a cobot first are payload, reach, and repeatability — and the worst practice is rating any one in isolation [S2]. A 5 kg payload cobot lifting a 4.5 kg gripper plus part is operating at 90% capacity, which kills cycle time and accelerates joint wear; spec ≥50% headroom over the worst-case mass including end-effector.
Reach is rated at the wrist center, not at the flange — subtract the gripper length and any vertical pedestal offset before accepting a model. Repeatability of ±0.02 mm is a typical premium-tier figure for 6-axis cobots; mid-tier units publish ±0.05 mm, and the calibration step documented in [S3] can add another 0.05 mm to that nominal band once the workcell is matched to the real production scene [S3]. Buyers comparing an articulated robot to a cobot at the same reach should not assume identical positioning performance, because cobot torque sensors and soft covers trade stiffness for compliance.
Safety Functions per ISO/TS 15066
Four collaborative functions defined in ISO/TS 15066 govern legal human-robot coexistence: safety-rated monitored stop, hand-guiding, speed-and-separation monitoring, and power-and-force-limiting — and the chosen cobot must document the one(s) it implements [S2][S3]. A power-and-force-limiting (PFL) arm clamps contact force below the biomechanical limits table; the operator's job becomes a clamp-current limit calculation against the worst-case moving mass and the protective stop response time.
Safety-rated monitored stop (SRMS) freezes all joints when a human enters the collaborative workspace; this function does not eliminate crushing risk and must be paired with a PFL mode or a hard guard for any tool that retains stored energy. For buyers reusing cells designed around traditional articulated robot guarding, the cheapest path is an SRMS cobot inside a light curtain envelope, but the cycle-time penalty of full stop-and-reset on every incursion often erases the ROI versus a PFL-rated arm.
Force, Speed, and Clamp Current Budget
Power-and-force-limiting selection is decided by the maximum quasi-static contact force the cobot can transmit to a human body region, not by the maximum joint torque [S2]. The governing equations in the published methodology weight payload, joint inertia, and protective stop reaction time; a slower stop response allows more kinetic energy to dissipate into the contact event, so the same arm looks safer on a cobot controller with a 50 ms stop than on a third-party safety relay with a 250 ms stop [S2].
Quoted maximum TCP speeds of 1 m/s are typical ceiling values for collaborative operation; derate to the actual collaborative mode limit once the risk assessment locks in. Buyers should reject any datasheet that publishes only the industrial-mode speed without a separate collaborative-mode speed entry, because the two differ by a factor of 2-5× on most platforms.
Integration Protocol and Cell Calibration
Ethernet/IP, PROFINET, EtherCAT, Modbus TCP and OPC UA are the five protocols a 2026 cobot cell is most likely to encounter; match the cobot's published fieldbus support to the existing PLC line before validating payload and reach [S2]. A cobot with a proprietary fieldbus-only stack adds a gateway cost and an extra failure node in the workcell, which works against the calibration fidelity Horváth et al. argue is necessary for resilient operation [S3].
Cell calibration — not arm calibration — is the single line item most often skipped by buyers. The Springer 2021 case study shows that nominal component models are mostly available from vendors, but matching them to the actual production scene (pedestal location, fixture datum, vision frame, conveyor encoder offset) requires measured-data fitting and carries its own residual error term [S3]. Budget at least 0.05 mm of extra tolerance for the calibrated workcell model on top of the cobot's published repeatability.
Selection Criteria vs Robot Type
The table below lines the three mainstream robot types a small-to-mid plant typically compares against a cobot — values are nominal mid-2026 platform ranges, not vendor-specific claims, and should be cross-checked against current datasheets. [S1]
Cobot (6-axis, PFL): payload 3-16 kg, reach 500-1300 mm, repeatability ±0.02-0.05 mm, collaborative mode speed 250-1000 mm/s, typical price band USD 25k-75k, ideal for mixed-product assembly, machine tending, screwdriving.
SCARA (4-axis): payload 1-20 kg, reach 200-800 mm, repeatability ±0.01-0.025 mm, no PFL mode (SRMS + guarding only), ideal for high-speed pick-and-place and small-part assembly with no human encroachment [S2].
Traditional articulated (6-axis, industrial): payload 5-800 kg, reach 700-3000+ mm, repeatability ±0.02-0.1 mm, no PFL, hard-guarded, ideal for heavy welding, palletizing, and high-inertia tasks where humans are excluded from the cell.
Who Should NOT Pick the Mainstream Cobot
High-speed palletizing, heavy welding (>10 kg wire-feed torches), and any application where the cycle demands more than 1 m/s sustained TCP speed are wrong fits for a PFL cobot and should be served by a guarded articulated or SCARA robot instead [S2]. The same logic applies to cleanroom and food-grade lines, where silicone cobot covers shed particles and complicate wash-down protocols — a stainless enclosed arm is the safer spec.
Mixed-product assembly with frequent changeovers, machine tending on CNC lathes with door interlocks, screwdriving with torque-controlled spindles, and pick-and-place alongside human packers are the four duty profiles where a PFL cobot pays back inside 18 months in the published selection cases [S2]. For line builders pairing the cobot with a remote I/O drop and a panel retrofit, the upstream remote I/O module selection guide is a useful sibling read for matching the fieldbus layer.
Shortlist Logic and Trackable Signals
A 2026 cobot shortlist should contain at least one model per major payload band (3-5 kg, 7-12 kg, 14-16 kg) from at least two vendors, each with a published ISO/TS 15066 compliance statement and a measured cell-calibration residual below 0.1 mm [S2][S3]. The decision should be made against a weighted AHP score, not on unit price, because the published methodology shows that integration cost and safety compliance cost routinely exceed the arm's purchase price over a 5-year horizon [S2].
Two signals to track before locking a vendor: the date of the most recent ISO/TS 15066 conformance test report on the vendor portal, and the release notes of the controller firmware (PFL clamp-current response time has tightened by 10-30 ms across two recent generation cycles). For buyers layering the cobot onto an existing HMI/SCADA stack, the HMI touch panel vs industrial display read pairs cleanly with the cell-calibration workflow described above [S3].